Note: move test models file in models to not import tests
directory, which breaks the mute_logger of test cases

* added intermediate ``_read_file`` step dispatching between CSV, ODS
  and XLS(X) parsing
  - primary dispatch on mime type, secondary on file extension:
    + OS may not provide a relevant mime type if no software locally
      installed for the filetype (e.g. Windows sends excel files as
      application/octet-stream if excel is not installed, and ODS files as
      zip if OpenOffice/LibreOffice isn't installed)
    + applications may re-register extensions to non-standard mimetypes
      breaking the dispatcher

    So if the mimetype is found trust it, otherwise try with the
    filename's extension (if any)
  - all readers skip lines with only empty cells in their output
  - ODS and XLS content are "CSVified", rows are converted to arrays of
    unicode strings
* UI altered to not assume CSV files everywhere, and avoid returning
  garbage preview data for non-CSV imports

Various:
* added a ``can_import`` utility function to tests.common, can be used
  to skip tests if an optional Python dependency is not installed (but
  one would like tests to run if the dependency is available)
* fixed datetime issue in ir_fields
* added converter for monetary fields (iso float)
* spreadsheet don't have integers, twiddling required to ensure integral
  values won't be serialized as floats (breaking conversion back to
  Python)
* improved some tests by asserting no error is generated (bonus: logs
  the error message if there is one, rather than just saying the result
  is blown)
* because some systems only provide elderly versions of
  XLRD (e.g. current debian stable provides 0.9.2 from April 2013)
  - keep using xldate_as_tuple instead of 0.9.3's xldate_as_datetime,
    workaround is simple
  - check for xlrd.xlsx in case of pre-0.8 XLRD

Authorship:
Ronak Baxi <rba@odoo.com>
Mohammed Shekha <msh@openerp.com>

Task 10792
Closes #7285